Doable General Equilibrium Models: Discussion

Abstract

The three papers are different yet have a common theme, the application of computable general equilibrium (CGE) modeling in agricultural policy analysis. Adelman and Robinson present a social accounting matrix (SAM) for the United States, emphasizing the agricultural sector. Derpanopoulos provides an optimal control formulation for CGE-like models. And lessons from experiences in applied CGE modeling are reviewed by Clarete and Roumasset. It is interesting and telling that the papers, although advocating the CGE approach, are about models only partially incorporating the associated concepts (Scarf). After some general comments on CGE modeling, brief observations are made on the three papers. Applications of CGE models in policy research raise a number of important questions on specification, estimation, solution approaches, and the selection of appropriate policy exercises. CGE models are static and have high prior information content. The separability and other assumptions on preferences and technology required to limit parameters and facilitate solutions are most plausible for more aggregate models. These aggregated specifications limit importantly the policy exercises that can be successfully undertaken with CGE models. Extensions to include dynamics in CGE-like structures are at present ad hoc and inconsistent with the CGE approach. There is interest presently in econometrically estimating CGE models (see Scarf and Shoven collection). But CGE models have high prior relative to empirical information content. If the empirical information content of the models is limited, then calibration and other estimation schemes that may not take best advantage of the data can be justified. Careful econometric estimation is most important for models with high empirical content. Since these models have high prior content, refinements in the way empirical information is introduced may have little impact on model outcomes. Arguments for calibration are better justified on this basis. Presently, it is possible to solve CGE models using readily available nonlinear programming algorithms. In fact, the dimensions of models solvable with the available technology are probably larger than can be justified, given the behavioral and technical underpinnings. Of course, currently available solution algorithms can be improved and extended to other general equilibrium models (Derpanopoulos). However, it is important to separate these extensions from the now-standard